A SYSTEMATIC ANALYSIS OF CATEGORY LEARNING: EFFECTS OF DIFFERENT LEARNING STRATEGIES AND TASK CHARACTERISTICS
A Master’s Thesis
by
ELİF CEMRE SOLMAZ
Department of Psychology İhsan Doğramacı Bilkent University
Ankara June 2019 ELİ F C EMR E S OL MAZ S YSTEMATI C AN AL Y S IS OF C AT EG ORY LEA R NI NG B il ke nt U ni ve rsity 2019
i
ii
A SYSTEMATIC ANALYSIS OF CATEGORY LEARNING: EFFECTS OF DIFFERENT LEARNING STRATEGIES AND TASK CHARACTERISTICS
The Graduate School of Economics and Social Sciences of
İhsan Doğramacı Bilkent University
by
ELİF CEMRE SOLMAZ
In Partial Fulfillment of the Requirements for the Degree of MASTER OF ARTS IN PSYCHOLOGY
THE DEPARTMENT OF PSYCHOLOGY İHSAN DOĞRAMACI BİLKENT UNIVERSITY
ANKARA June 2019
iv ABSTRACT
SYSTEMATIC ANALYSIS OF CATEGORY LEARNING Solmaz, Elif Cemre
M.A., Department of Psychology Supervisor: Asst. Prof. Dr. Miri Besken
June 2019
Many studies showed that organization of study materials has a strong effect on learning performance (Kornell & Bjork, 2008). The current study compared category learning performance through blocked and interleaved learning conditions, using verbal
(Experiment 1) and pictorial (Experiment 2) materials. Participants were assigned to one of three conditions at encoding phase: blocked, interleaved, and semi-interleaved
learning conditions. In blocked learning, participants studied four exemplars of the same category within the same trial, and categories were blocked across trials. In interleaved learning, participants studied one exemplar from four different categories within the same trial, and categories were interleaved across trials. In semi-interleaved learning, participants studied four exemplars of the same category within the trial, but categories
v
were interleaved across trials. At testing phase, participants were tested on old and novel exemplars of the categories that they studied and were asked to identify the category of each exemplar. Lastly, they were tested on explicit understanding of categories. Both Experiment 1 and 2 revealed that participants produced the highest learning performance in the semi- interleaved learning condition. Learning similarities within the same trial and learning differences across trials might lead to the most optimal learning strategy for category learning, regardless of the type of stimuli used.
Keywords: Categorization, Effective Learning Strategies, Induction, Memory, Spacing
vi ÖZET
KATEGORİ ÖĞRENİMİNİN SİSTEMATİK ANALİZİ
Solmaz, Elif Cemre Yüksek Lisans, Psikoloji Bölümü
Tez Danışmanı: Dr. Öğr. Üyesi Miri Besken Haziran 2019
Çalışılan materyallerin nasıl organize edildiğinin öğrenme üzerindeki etkisi birçok çalışma tarafından gösterilmiştir (Kornell & Bjork, 2008). Bu çalışma, dönüşümlü öğrenme (interleaved learning) ve blok öğrenme (blocked learning) yöntemlerini sözsel (Deney 1) ve görsel (Deney 2) materyaller kullanarak karşılaştırmıştır. Çalışmanın ilk aşamasında katılımcılar blok, dönüşümlü ve yarı dönüşümlü olmak üzere üç çalışma koşulundan birine atanmışlardır. Blok öğrenme koşulunda, katılımcılar her çalışma denemesinde bir kategorinin dört örneğini gördüler ve denemeler arasında da kategori örnekleri sırayla sunuldu. Dönüşümlü öğrenme koşulunda, katılımcılar her deneme ekranında dört kategorinin birer örneğini gördüler ve örnekler arasında da kategoriler dönüşümlü olarak gösterildi. Yarı dönüşümlü öğrenme koşulunda ise, katılımcılar
vii
dönüşümlü olarak gösterildi. Test aşamasında katılımcılar çalıştıkları ve çalışmadıkları örnekler için test edildiler. Deneyin son kısmında kategorilerle ilgili açık öğrenme (explicit learning) konusunda test edildiler. Deney 1 ve Deney 2’nin sonuçları yarı dönüşümlü öğrenme yönteminin en yüksek öğrenme performansı sağladığını
göstermiştir. Yarı dönüşümlü öğrenme yöntemi, bir kategori içindeki benzerliklerin ve kategoriler arasındaki farklılıkların daha kolay fark edilmesini sağlamaktadır. Bu nedenle, yarı dönüşümlü öğrenme yöntemi blok ve dönüşümlü öğrenme yöntemlerine göre daha başarılı bir kategori öğrenimi sağlamaktadır.
Anahtar Kelimeler: Bariz Öğrenme, Bellek, Etkili Öğrenme, Kategori Öğrenimi
viii
ACKNOWLEDGMENTS
First thing first, I would like to express my gratitude with most sincere honesty to Dr. Miri Besken. I have learnt everything I know from her in terms of academic life for four years. She will always be more than an advisor for me. In addition, I express my
appreciation to Dr. Gül Günaydın for her support who teaches courses that I will never forget during my life.
I would also like to express gratitude to my mother, father, my lovely sisters Ceren and Ceyda for always believing in me and encouraging me.
I owe to special thanks to Öykü Çiftçi for being my biggest supporter, best roommate and best colleague during my master life. I am also grateful to Dilay Seda Özgen for being my roommate and being there whenever I need. I would like to thank my VIP team who are Hanzade Tepeoğlu, Doğukan Kaya, Durukan Güven, Berka Tanyeli, Beyza Özgöde, Mehmet Günçavdı and Eren Bilaloğlu. They made my Bilkent life unforgettable.
I express my appreciation to my colleagues Irmak Tuğcu and Cansu Gökçe who we started this psychology journey together.
I owe to special thanks for all their support to Burcu Seydioğlu, Okan Ok and
Abdülkerim Satuk Buğra İsmailefendioğlu who have been with me since the first day of my Bilkent life.
ix
I express my appreciation to my childhood friends; Dilara Yılmaz, Elif Kürüm, Havva Zeynep Koç, Emine Koç who are the people who made me the person that I am.
I would like to thank to all, Ecem Eylül Ardıç, Gamze Nur Eroğlu & Ezgi Melisa Yüksel for their emotional support. I also would like to thank to all of the undergraduate
students of Cognitive Psychology Lab for helping me with my thesis.
Lastly, I also would like to thank myself for bringing this process to the end with my patience and effort.
x TABLE OF CONTENTS ABSTRACT ... iv ÖZET ... vi ACKNOWLEDGMENTS ... viii TABLE OF CONTENTS ... x
LIST OF TABLES ... xii
LISTS OF FIGURES ... xiii
CHAPTER I: INTRODUCTION ... 1
1.1 Effective Learning Strategies and Category Learning ... 2
1.2 Research that Found Evidence for Superiority of Interleaved Learning Strategy ... 5
1.3 Research that Found Evidence for Superiority of Blocked Learning Strategy ... 9
1.4 Research that Found Evidence for Both Interleaved and Blocked Learning Strategies ... 11
1.5 Present Study ... 14
CHAPTER II: EXPERIMENT 1 ... 19
2.1 Method... 19
2.1.1 Participants ... 19
2.1.2 Material and Design ... 19
2.1.3 Procedure ... 22
xi
CHAPTER III: EXPERIMENT 2 ... 34
3.1 Method... 34
3.1.1 Participants ... 34
3.1.2 Material, Design and Procedure ... 34
3.2 Results ... 34
CHAPTER IV: COMPARISON OF EXPERIMENT 1 & EXPERIMENT 2 ... 42
4.1 Results ... 42
CHAPTER V: GENERAL DISCUSSION ... 49
REFERENCES... 59
APPENDICES ... 62
APPENDIX A. CATEGORIES AND EXEMPLARS ... 62
xii
LIST OF TABLES
Table 1. Findings Of Previous Research ... 13 Table 2.Reaction Times For First And Second Presentations Of Exemplars In Learning Phase ... 28 Table 3. Mean Accuracy Rates For Old And New Exemplars In Exemplar Testing Phase And The Standard Deviations In Parentheses ... 30 Table 4.Unconditonal Response Times For New And Old Exemplars In Exemplar Testing Part ... 31 Table 5.Conditonal Response Times For New And Old Exemplars In Exemplar Testing Part ... 32 Table 6.Inference Rate For Actual Category Names ... 33 Table 7. Reaction Times For First And Second Presentations Of Exemplars In Learning Phase ... 35 Table 8. Mean Accuracy Rates For Old And New Exemplars In Exemplar Testing Phase ... 37 Table 9.Unconditional Response Times For New And Old Exemplars ... 38 Table 10.Conditional Response Times For New And Old Exemplars In Exemplar
Testing Part ... 40 Table 11.Inference Rate For Actual Category Names ... 41
xiii
LISTS OF FIGURES
Figure 1. Sample For Blocked Learning Condition ... 23
Figure 2. Sample For Interleaved Learning Condition ... 23
Figure 3. Sample For Semi-Interleaved Learning Condition ... 24
Figure 4. Sample For Testing Trial ... 26
Figure 5. Reaction Times Of Experiment 1 And Experiment 2 For All Learning Conditions ... 43
Figure 6. Testing Accuracy Of Old And New Exemplars For Experiment 1 And Experiment 2 ... 44
Figure 7.Unconditional Response Times Of Old And New Exemplars Of Experiment 1 And Experiment 2 ... 45
Figure 8.Conditional Response Times Of Old And New Exemplars Of Experiment 1 And Experiment 2 ... 46
1
CHAPTER I
INTRODUCTION
Which learning strategy produces better learning performance is a debated issue for a long time. One of the controversial issues is interleaving vs. blocking of study materials. Blocked learning strategy refers to a study method, in which one studies the same type of materials consecutively before moving onto the next type of material. For example, when there are three items (A, B and C) to study, blocking learning strategy would require one to study the materials in the following format: A1 A2 A3 B1 B2 B3 C1 C2 C3. On the other hand, interleaved learning strategy refers to studying materials in a mixed order. An example of this format is as follows: A1 B1 C1 A2 B2 C2 A3 B3 C3 (Kurtz & Hovland,1956). Previous research generally shows that students do not choose effective learning strategies during their study sessions. For instance, although most of the
research showed that interleaving study materials produces higher memory performance than blocking them, most of the students believe that blocking is a better learning strategy than interleaving (Cohen, Yan, Halamish & Bjork, 2013). On the other hand, students may not be choosing a specific learning strategy all the time, because there have been contradictory findings in literature. Some studies find that interleaved studying produces higher learning (Birnbaum, Kornell, Bjork & Bjork, 2012; Kornell & Bjork,
2
2008), whereas some others find that blocked learning leads to better memory performance (Carpenter & Mueller, 2013; Sorensen & Woltz, 2016).
The first aim of the current study is to contribute to this literature through manipulating study order, using blocked, interleaved and semi-interleaved learning strategies.
Previous research also reveals that interleaved learning strategy generally produces superior learning for pictorial materials, whereas blocked learning strategy is associated with superior learning of verbal materials. Therefore, the second aim of the current study is to investigate how nature of the study materials (words vs. pictures) affect learning performance.
1.1 Effective Learning Strategies and Category Learning
How to study effectively has been a very controversial issue among students for many years. A branch of educational psychology investigates effectiveness of different
learning strategies (Rohrer & Taylor, 2007; Mitchell, Nash, & Hall, 2008; Olina, Reiser, Huang, Lim & Park, 2006). Studies in this area have shown that almost every student uses a different strategy to study. Because different study strategies provide different levels of learning performance, some students are more successful than others. Thus, some students use more effective learning strategies naturally, whereas a majority of them are not aware of their ineffective learning strategies (Hartwig & Dunlosky, 2012). One of the most investigated learning strategies is spacing of study materials. Spacing is a widely investigated issue, because the organization of study materials like blocking or interleaving have a strong effect on learning performance (Kornell & Bjork, 2008). In blocked learning strategy, stimuli that belong to the same concepts are studied
3
concepts are studied together that they are intermixed together (Kurtz & Hovland, 1956). Interesting point in spacing (vs blocking) study strategy is that generally students cannot accurately evaluate which learning strategy is more effective. Thus, there is a
discrepancy between people’s metacognitive judgments for learning strategies and their actual memory performance. Previous research showed that people’s beliefs about effective learning strategies are against some scientific findings. For instance, Cohen et al. (2013) showed that participants choose to block more often than interleaving
although interleaved learning strategy provides better learning performance in most cases. In addition, Kornell & Bjork (2008) and Hartwig & Dunlosky (2012) replicated this finding. Therefore, finding which learning strategy produces the most optimal learning performance is important in terms of raising students’ awareness.
The effectiveness of blocked and interleaved learning strategies has been measured through different materials, such as mathematical problems (Rohrer et al, 2009), words (Carpenter & Mueller, 2013; Sorensen & Woltz, 2016) and visual materials (Kornell & Bjork, 2008; Birnbaum, Kornell, Bjork & Bjork, 2012). Most of those studies use study materials through creating categories, because when participants see categories, they have to identify differences between categories and shared characteristics within
categories. Sana, Yan & Kim (2017) also argues that boundaries of category and features that differ one category from another category should be learnt successfully when
separate categories are being learned. Understanding commonalities that are shared by all exemplars of a category is also an important point in category learning. Therefore, while studying categories, people should make inferences about given categories. Likewise, with category learning, participants’ explicit understanding about stimuli can
4
be measured because they should apply the rules to novel exemplars (exemplars belong to same categories that they do not study during encoding phase) that they inferred from studied exemplars. In addition, Carvalho & Goldstone (2014) argues that in category learning, how information is organized has huge impact on what is successfully learnt. Therefore, category learning can provide useful information about which learning strategy provides better memory performance than other learning strategies.
One of the initial studies that investigated effectiveness of interleaving vs. blocking was conducted by Kurtz & Hovland (1956). They used geometric shapes as category
materials that differed in color, shape, size and position. During encoding phase,
experimenters showed four stimuli to participants and they were asked to study different type of exemplars either in interleaved or blocked learning strategies. After encoding phase, participants were asked to report common features of each category verbally. If participants gave wrong answers, experimenters repeated the trials until participants reported the true categorization features. Then, participants were tested with an
immediate recognition test, in which they were presented with four different exemplars and asked to categorize exemplars. Results revealed that blocked learning produced better category learning than interleaved learning condition. In recent studies, procedure is slightly different than past studies. The procedure of more recent studies consists of three main phases: encoding, distraction and testing phases. During the encoding phase, participants are presented with a stimulus from one category and they are asked to classify presented stimulus. It is followed by feedback for each response they give. The learning strategy is usually manipulated by the experimenter. Participants may be exposed to blocked or interleaved studying strategy in a within or between subject
5
design. In the second part, participants are asked to do some trivia task to distract their attention for two or three minutes. The last phase is the testing phase. During the testing phase, participants are tested on implicit and explicit understanding about categories. Implicit understanding is generally tested by testing studied exemplars. Explicit
understanding is tested through using novel exemplars and also free recall test questions about the characteristics of the previously studied categories.
1.2 Research that Found Evidence for Superiority of Interleaved Learning Strategy Like students, experimenters that investigate effectiveness of interleaved and blocked learning also have not come to a consensus yet. There are contradictory results about effectiveness of blocking and interleaving. There are many studies that found support for the superiority of interleaved learning strategy over blocked learning. For instance, as one of the initials studied that revealed higher learning performance for interleaved learning condition was conducted by Kornell and Bjork (2008). They tested blocked and interleaved learning strategies by using category learning. They used six paintings from twelve different artists as categories. In Experiment 1A, participants studied items both in blocked and interleaved learning conditions in a within subjects design. In Experiment 1B, participants were assigned to either blocked or interleaved learning conditions in a between subjects design. Consequently, all participants were tested on novel exemplars in testing phase. The results of the study revealed that participants’ performance was better in interleaved learning condition than blocked condition.
There are a large number of other research that support advantage of interleaved learning strategy. One of the most known studies that was conducted by Kang and Pashler (2012) in which they also used paintings of Kornell & Bjork (2008) as category materials. They
6
conducted two experiments. In Experiment 1, Kang and Pasher (2012) used temporally spaced learning condition in addition to massed and interleaved conditions. In the temporally spaced leaning condition, there were ten seconds delay between exemplars. Moreover, they added another learning condition in which they provided participants four paintings of the same artist at the same time rather than displaying each painting on the screen sequentially, referred to as simultaneous massed condition. Experiment 1 revealed that interleaved condition produced higher learning performance at test than all other three conditions. In Experiment 2, they tested the difference between massed, interleaved and simultaneous different conditions, in which participants were presented with four different paintings from four different artists on the same screen. Experiment 2 showed that simultaneous different condition produced higher learning than interleaved condition. In addition, interleaved condition produced higher learning than massed condition.
In another study, Rohrer, Dedrick & Burgess (2014) used mathematics problems as material in their study, because they claimed that generally in the math problems, same kind of problems and strategies are taught to children and while they solve same kind of problem at the same time which is also known as blocked learning, they do not need to think about the problem and they just apply the rule to question. Thus, they tested 7th -grade students about math problems which were presented in blocked and interleaved condition in a within-subject design. Results of the study showed that interleaved learning provided an average performance twice as high as blocked learning. Similar to Rohrer et al (2014) study, Taylor and Rohrer (2010) investigated spacing effect with
7
using math problems. Results showed superiority effect of interleaved learning condition.
In addition to intervening of different stimuli, Birnbaum et al. (2012) argued that interleaving provides superior memory performance than blocked learning, not because of the temporal spacing, but because of that participants can grasp the different features of categories in interleaved learning more easily than blocked learning condition, because exemplars of different categories intervene with each other. They conducted an experiment, using different bird and butterfly family exemplars. In Experiment 1, there were three conditions which measured the difference between temporal spacing and alternating between stimuli; contiguous (no spacing between exemplars), alternating trivia (irrelevant questions before every photo) and grouped-trivia (eight photos then eight trivia questions). All participants studied exemplars in one of these interleaved learning conditions. Results showed that the accuracy of the participants was higher in the grouped trivia condition as compared to alternating-trivia and contiguous conditions. Thus, interleaving the stimuli provided better memory performance than temporal spacing. In Experiment 2, participants studied items in blocked and interleaved
conditions, but they were either in contiguous or temporally- spaced (delayed -10s- then trivia questions) conditions. Results revealed that interleaved learning conditions
produced superior category learning than blocked learning conditions. In contrast, in temporally spaced condition, there was no significant difference between blocked and interleaved learning strategies. Therefore, they contended that the effect of interleaving is not related to time. Spacing effect occurs, because exemplars intervene with each other, which allows participants to compare and contrast the exemplars from different
8
categories more efficiently. In a similar line of work, Zulkiply and Burt (2013) conducted a study, in which they used Kornell and Bjork’s (2008) paintings. In the experiment, there were four conditions which were blocked-immediate, interleaved-immediate, blocked-temporally spaced and interleaved-temporally spaced learning conditions. During blocked-temporally spaced and interleaved temporally-spaced conditions, presentation of each exemplar were followed by thirty seconds unrelated task. Through these conditions, they tried to find difference between temporally spaced and interleaving study strategies. Results showed that interleaved-immediate and interleaved-temporally spaced conditions enhanced memory more than
blocked-immediate and blocked-temporally spaced conditions. Related study, Zulkiply, McLean, Burt & Bath (2012) also used psychopathological disorders. Participants studied
different disorder cases and then they were tested on new cases and Zulkiply et al. (2012) found that interleaved learning provided better induction than blocked learning for understanding different psychopathological disorders. This study also showed importance of interleaving for category learning.
Studies which produce more efficient and higher performance for interleaved learning strategy than blocked learning have been explained through
discriminative-contrast-hypothesis, which is also known as spacing effect. Discriminative-contrast-hypothesis
claims that when studied stimuli intervene with each other, it provides better learning performance than blocked learning (Cepeda, Pashler, Vul, Wixted & Rohrer, 2006).
Discriminative-contrast-hypothesis explains superiority of interleaved learning strategy
with that when participants see one exemplar from different categories, they can distinguish differences between categories easier than blocked learning strategy. Since
9
participants can distinguish between category differences more easily than blocked learning strategy, it provides more induction than blocking (Carvalho & Goldstone, 2014). Sana et al. (2017) claimed that interleaved learning strategy provides higher memory performance than blocked learning strategy, because in interleaved learning participants see different exemplars in mixed order, which leads to discrimination of salient features of categories more easily than blocked learning category. In blocked learning, understanding salient features of categories is harder. For instance, Taylor and Rohrer (2010) conducted a study which explains superiority of interleaved learning strategy. They asked participants solve mathematical questions which they had to make calculation about four dimensions (edge, face, angle and corner) of prisms. Participants were assigned to interleaved or blocked learning conditions and results revealed that interleaved learning strategy provided better problem solving than blocked learning strategy. In addition, Taylor and Rohrer reported that participants in blocked learning strategy made more discrimination errors than interleaved learning strategy while
solving problems in testing phase. Thus, discriminative-contrast-hypothesis supports the superiority of interleaved learning strategy over blocked learning strategy, because of its ability to provide easy discrimination between different features of different category members leading to more efficient category discrimination.
1.3 Research that Found Evidence for Superiority of Blocked Learning Strategy
However, spacing effect is not applicable for every circumstance. In contrast to spacing
effect, there are many other studies that found advantage of blocked learning strategy.
Foss (1968) conducted a study by using linguistic systems decades ago. Results revealed that blocked learning produced higher learning performance than interleaving of study
10
materials. A recent study conducted by Sorensen and Woltz (2016) also found evidence for superiority of blocked over interleaved learning strategy. They used verbal category materials and generated four conditions which were high blocked, medium blocked, low blocked and all interleaved. In the testing phase, participants were tested for studied and novel exemplars. At the end of the experiment, they were asked to predict the actual category names for the categories with the artificial names that they studied at the beginning of the experiment. Results of the experiment showed that in all-interleaved condition participants made more errors than other conditions. Also, all-interleaved condition produced lower learning performance than other conditions, specifically for novel category exemplars, revealing that participants in the all-interleaved condition had lower general understanding of exemplars and actual category names. Therefore, this study showed that interleaved learning condition may sometimes produce lower performance, compared to blocked learning condition when verbal materials are used. In a similar line of work, Carpenter and Mueller (2013) tested blocked and interleaved learning strategies, using French vocabulary as their study material. They conducted four experiments that included both within and between subject designs, multiple choice test (recognition) and recall test. In addition, they manipulated level of exposure to stimuli during experiments. Participants were presented four exemplar words for eight different pronunciation rules. Participants were native English speakers and they were asked to study the pronunciation of some French words. Results of the study revealed that in all conditions and all experiments blocked learning strategy provided better word
11
Studies that found superiority effect for blocked learning strategy usually explain the findings through massing-aids-induction-hypothesis. Massing-aids-induction-hypothesis claims that blocked learning strategy provides higher category learning than interleaved learning because participants can easily discriminate shared characteristics of exemplars belonging to the same category. Therefore, massing-aids-induction-hypothesis argues that blocked learning condition enables comparison within categories more efficiently than interleaved learning strategy. People can realize common features of the category easily, therefore they can make abstraction about what a category represents more easily than interleaved learning strategy (Sana et al., 2017).
1.4 Research that Found Evidence for Both Interleaved and Blocked Learning Strategies It is important to emphasize that not all studies produce a definitive finding for the superiority of one learning strategy over the other for interleaved or blocked learning. Some of the studies that used slightly different methodology find evidence for
conditions in which either of the learning strategies may be more effective. For instance, level of similarity between and within categories may sometimes change which learning strategy will be more effective for category learning. For instance, Carvalho and
Goldstone (2014) conducted a study to test interleaved vs. blocked learning strategies. They hypothesized that superiority of interleaved learning condition may originate from different levels of similarity between the presented categories and also because of the delayed study opportunity of interleaved study. In experiment 1, they used “Blobs” which are some artificial shapes. They manipulated similarity in categories (low vs. high) in a between-subjects design. They also manipulated study condition (interleaved vs. blocked) and test type (immediate vs. delayed) in a within-subjects design. Results
12
revealed that in overall, blocked learning was more beneficial than interleaved study, low similarity better than high similarity. In addition, those participants who were presented with low similarity categories improved more in category learning than participants who were presented with high similarity in interleaved learning strategy better than blocked learning strategy.
Sana et al. (2017) also tested effect of interleaving (vs. blocking) by using statistic chapters. They found that when stimuli were juxtaposed in mixed order, interleaved learning strategy produced higher learning performance than blocked study (Experiment 1), but when stimuli were spaced temporally, spacing effect reduced (Experiment 2). They found that when stimuli were presented three-at-a-time rather than one-at-a-time, also blocked learning strategy provided more induction than interleaved learning. This is consistent with massing-aids-induction-hypothesis, because when categories are presented three-at-a-time, shared characteristics of the same category are figured out easier. In a similar line of work, Pan, Tajran, Lovelett, Osuna and Rickard (2018) found that using blocked or interleaved learning strategy may benefit learners at different rates, depending on if the material is repeated or presented only once. They presented Spanish words English speakers to teach them how to conjugate. When stimuli were presented multiple times, interleaving was more beneficial for learning over massed learning. In contrast, when stimuli were presented with no repetition, blocked learning strategy provided superior learning performance.
As mentioned in the foregoing literature review, Zulkiply and Burt (2013) found that interleaving the stimuli helps category induction when the discriminability of categories is low (Experiment 2). They used categories that consisted of different kinds of shapes
13
color and different levels of discriminability as low or high discriminability. Results showed when discriminability was high between categories, blocked condition provided higher learning than interleaved condition. As the massing-aid-induction-hypothesis claims, blocked learning strategy helped participants to abstract the features within the same category more easily. In addition, because within category similarity was high, blocking produced more learning than interleaving.
Table 1. Findings of previous research
Study Material Findings
Kornell & Bjork (2008) Paintings Interleaved > Blocked Kang & Pashler (2012) Paintings Interleaved > Blocked Birnbaum et al. (2013) Birds & Butterflies Interleaved > Blocked Carpenter & Mueller (2013) French Words Blocked > Interleaved Sorensen & Woltz (2016) Words Blocked > Interleaved Sana et al. (2017) Statistic Chapters / 1-at-a-time
Statistic Chapter / 3-at-a-time
Interleaved > Blocked Blocked > Interleaved
Pan et al. (2018)
Spanish Words w/ repetition Spanish Words w/ no repetition
Interleaved > Blocked Blocked > Interleaved
Besides blocking and interleaving of studying, people’s mental effort is also effective about impact of learning strategies. Carvalho and Goldstone (2015) conducted a study to investigate the attentional mechanisms behind interleaved and blocked learning. To study attentional mechanisms, they used passive versus active learning method in their
14
experiments. In passive learning, participants are provided with all information about stimuli. On the other hand, in active learning, participants are provided with parts of the information about stimuli and they are expected to make inferences about the given stimuli. In Experiment 1, they used different kinds of Fribbles, artificial objects which look like animals with different extensions that make them belong to separate categories. Participants were assigned either passive learning condition or active learning condition. Experiment 1 produced no main effect for presentation of stimuli (blocked vs
interleaved), type of item (studied vs novel) and study group (passive vs. active), but they found an interaction between presentation of stimuli and study group. In active learning condition, interleaved learning produced higher category learning than blocked learning condition, whereas, in passive learning condition, blocked learning produced superior performance than interleaved condition. In Experiment 2, they used categories which had high similarity between categories but low similarity within categories. Experiment 2 replicated the results of Experiment 1. Thus, the superiority of the learning strategy may also change, depending on the level of activity of participants during encoding.
1.5 Present Study
The first aim of the present study is to investigate effectiveness of spacing (vs blocking). Therefore, participants were presented with materials in three different learning
strategies: blocked, interleaved and semi-interleaved. In blocked learning strategy, participants are presented with stimuli consecutively, whereas in interleaved learning strategy are presented stimuli by mixed together. In addition, semi-interleaved learning strategy includes a hybrid combination of both blocked and interleaved learning
15
strategies which are mentioned in detail in the method section. Consistent with previous experiments, current experiment also made use of categories as material. The method used in the current experiments was similar to that of Sorensen and Woltz (2016). Sorensen and Woltz presented participants words in six separate conceptual categories. For example, participants could be presented with exemplars of a category ‘something that is loud’, with eight exemplars in each category. “Something that is loud” category included “trumpet, thunder, cymbals, siren, jackhammer, shout, scream and dynamite”. Additionally, exemplars were presented with non-word category names instead of actual category names name such as “HEECE” for “something that is loud”. Thus, they were expected to make inferences for actual category names.
Experiment of Sorensen and Woltz (2016) consisted of three parts which were encoding, distraction and testing phases. During encoding phase, participants assigned to
interleaved learning condition which they were presented with one exemplar from four different categories with non-word category names in each trial. On the other hand, participants assigned to blocked learning condition which they were presented with four exemplars of the same category in one trial. Participants were asked to study for
exemplars and try to guess actual category names for the testing phase. Distraction phase consisted of letter sequencing exercises. Testing phase included two parts: exemplar testing and category testing. In exemplar testing, a recognition test for studied and novel exemplars was used. For explicit category testing, participants were asked explicitly what each non-word category name symbolized.
The present study is a modification on Sorenson and Woltz (2016). Consistent with method of Sorensen and Woltz (2016), twelve categories were formed for encoding part
16
of present experiment. Each category consisted of eight exemplars. Categories were formed as abstract categories which means that they were not natural categories like bird families. For instance, in the current study, one category was “something that is white”, including the exemplars “cloud, whiteboard, wedding dress, swan, milk, teeth, cotton, polar bear”. Therefore, categories did not have natural bonding, but exemplars of each category shared some commonality. In addition, non-word category names for each category were formed and categories were presented with those non-word category names rather than actual category names. To test implicit learning performance of participants, they were tested on studied exemplars and novel exemplars during testing phase. Finally, participants were asked to make predictions about actual category names of categories that they studied during encoding phase. Participants’ performance on novel exemplars testing and also predictions show that which learning strategy provides better explicit learning.
With three learning conditions, the present research aims to test
induction-hypothesis and discriminative-contrast-induction-hypothesis. According to
massing-aids-induction-hypothesis, blocked learning strategy in both experiments should produce higher learning performance at test than the other two conditions, because participants should understand within category similarities better than others. Therefore, they should have an abstraction for the presented category at the end of studying each category. On the other hand, according to discriminative-contrast-hypothesis, interleaved learning strategy should provide better learning performance than other conditions, because participants should discriminate differences between categories easier because they see four exemplars from different categories in the same trial. The current research
17
hypothesized that semi-interleaved learning strategy which includes both blocked and interleaved learning conditions should produce the best performance compared to other conditions, because semi-interleaved learning conditions can benefit from both within category abstraction of blocked learning and between category discrimination of interleaved learning. Therefore, participants should easily understand what common point of presented category and also what differentiates the presented category from the next category that it is interleaved with.
As can be inferred from the foregoing literature review, generally studies that used pictorial material found superiority effect for interleaved learning strategy over blocked learning strategy (Kornell & Bjork, 2008, Zulkiply et al., 2011; Zulkiply & Burt, 2013; Rohrer et al, 2012; Birnbaum et al., 2012), whereas, studies that used words as material found superiority effect for blocked over interleaved learning strategy (Carpenter & Mueller, 2013; Sorensen & Woltz, 2016; Pan et al., 2018b) (See Table1). Like mentioned in research that found evidence for both interleaved learning and blocked learning strategy, when materials or procedure of experiments are changed, benefit of interleaved and blocked learning strategies may also change. Therefore, the present study is conducted for systematic comparison of verbal and pictorial materials in different learning conditions. Although there are many studies used pictorial and verbal material, to our knowledge, there are no studies that directly investigated the difference between words and pictures in the literature. In the current study, I presented participants with words in Experiment 1 and pictures in Experiment 2. I hypothesized that if learning conditions affect learning performance independent from what is studied, there should not be learning performance difference between Experiment 1 and Experiment 2. On the
18
other hand, if Experiment 1 and Experiment 2 will produce significantly different results, it can be explained by nature of study materials
19
CHAPTER II
EXPERIMENT 1
2.1 Method 2.1.1 Participants
Based on a power analysis, a hundred and two participants were required to detect a large effect size f of .42 to have a 95% power when the traditional criterion of .05 for statistical significance is employed. Consistent with power analysis, a hundred and two Bilkent University students, thirty-four for each condition, who are between the ages of 18-30 and native speakers of Turkish participated in the study. Participants were
compensated with course credit. All participants were randomly assigned to one of three conditions.
2.1.2 Material and Design
Twelve artificial categories were created for the experiment. Eight exemplars were determined for each category. Artificial categories refer to exemplars which do not have natural bonding. I used these artificially-formed categories instead of natural categories, in which exemplars had natural commonalities, because using naturally bonded
categories would make the procedure easier which can create a confound. When participants understood categories easier because of the nature of the study, it can moderate effect of learning strategies. Moreover, since the study is a modification of Sorenson and Woltz (2016), the materials were kept as similar to the original as
20
possible, as they also used artificial categories. Four of the exemplars were presented in the encoding phase and the other four exemplars were used in the testing phase as novel exemplars. Presentation of exemplars was counterbalanced across participants.
Exemplars of each category were unique to the determined category, such that they could not be categorized under the description of any of the other categories. For instance, “cloud” from “something that is white” category did not fit to “something that is yellow” or “something that is made of steel” or “something that is made of wooden” (See Appendix A for further details). Also, a non-word category name for each category was formed. Each non-word category name started with a different letter of the alphabet and none of these names had a meaning in Turkish (See Appendix A1 for all non-word category names). Exemplars were presented under those non-word category names during the experiment.
The experiment had a between-subjects design. Participants were assigned to one of the three learning conditions: blocked learning, semi-interleaved learning and interleaved learning. All conditions had three study blocks during the encoding phase. Each block included 4 categories and 4 exemplars from each category. Each exemplar was presented twice during the learning phase. Therefore, each block had 32 trials. In the blocked condition, all categories were presented in order. After each exemplar of one category was presented, participants were asked to study next category and its exemplars. Thus, in blocked learning condition participants saw 4 exemplars of one category within the same display. Then, they moved onto the following category sequentially. In the interleaved learning condition, all categories were distributed randomly within each block. In each trial of interleaved learning condition, participants were presented with
21
four different categories and one of those category exemplars within the same display. Presentation of all categories and their exemplars were randomized across participants. In the semi-interleaved condition, participants were presented one blocked category for the first scene and another blocked category for second scene. Therefore, different categories were presented for each scene. Also, order of categories was randomized across participants.
Two separate lists were created to counterbalance for the study status at test across participants. The first list consisted of 48 words, which were the same words in encoding phase as old exemplars and other lists consists of 48 words belonging to same categories as new exemplars. For example, if the word “cotton”, “polar bear”, “milk", “wedding dress" were presented during the learning phase to half of the participants within category “something that is white”, the rest of the category (teeth, swan, cloud and whiteboard) was only presented during the category inference test. The items that are presented during learning phase are “old” items, whereas the items that have not been presented before at learning phase are “new” items. The study status (old vs. new) was counterbalanced across participants, and these lists were presented to an equal number of participants in each condition.
During the encoding phase, reaction times for first and second presentation for each response to each stimulus were recorded. In the encoding phase, to be consistent with Sorensen & Woltz (2016) that procedure of current research modified by their method, each exemplar were presented for 6 seconds. Similarly, during the testing phase, reaction times for response to old and new exemplars were recorded by a computer program.
22
Finally, category lists were prepared to measure participants’ inferences about relations between exemplars and categories. Those lists included all 12 non-word category names. 2.1.3 Procedure
Participants were tested individually on desktop computers and standard keyboard with an Intel® Core™ processor in the Bilkent University Cognitive Laboratory Area. They were tested through E-prime® software program. The experiment consisted of four parts: encoding, distraction, exemplar testing and category testing. Participants were given different instructions for each learning condition. Participants assigned to blocked condition were instructed that they would be presented with four exemplars of one category sequentially. After all exemplars of one category were presented twice in a row, the program would proceed onto the other category exemplars in the same block. In interleaved learning condition, participants were instructed that they were going to see one exemplar from 4 categories in one trial and exemplars would be changed in each trial. Also, in semi-interleaved condition, participants were told that they would see one category and its four exemplars in one trial and in next trial they would see another category with its four exemplars.
23
Figure 1. Sample for blocked learning condition
24
Figure 3. Sample for semi-interleaved learning condition
In the learning phase of the experiment, participants were presented four exemplars with category names at top of the screen. In the middle of the screen, one of exemplars that was presented at the top of the screen was presented as target word. Also, at the bottom of the target word, two letters were presented. One letter was first letter of correct category name and other letter was first letter of one of the other category names. Therefore, participants were asked to press one of the letters from keyboard in 6
seconds. When they pressed one of the keys, feedback was provided about the accuracy of their responses. If they took more than 6 seconds, the program warned them to respond faster. In all conditions, each exemplar was presented two times during
encoding phase. Moreover, in the encoding phase of blocked condition, 4 exemplars of same category were presented in one trial. In the blocked condition four same non-word category names were seen at the top of scene (See Figure1). Under each category names, exemplars of the category were presented. For instance, in the category of “something is white”; polar bear, milk, cloud and cotton were presented in that trial. In
25
blocked learning condition exemplars of categories were studied one after another. After four exemplars of one category were studied, program proceeded onto another category. In the interleaved condition, 4 different categories and one exemplar from the 4 different categories were presented in one trial. For instance, “something that is white”,
“something that is yellow”, “something that is made of wooden” and “something that is made of steel” were presented under with one exemplar for each different category in one trial (See Figure2). Other procedure was the same as blocked condition. Exemplars and categories were randomized for each block. In the semi-interleaved condition, trials were the same as blocked conditions, but order of the trials were different (See Figure3). Therefore, in the first trial, participants saw “something that is white” category, in second trial, participants saw “something that is yellow category”, in third trial, they saw “something that is made of wood” and in the fourth trial, they saw “something that is made of steel”. Presentation of trials were the same as blocked conditions. Once participants studied all the items within each category, they proceeded onto the
distraction phase. Distraction phase was same for all conditions. In the distraction phase, participants were asked to solve arithmetic problems for 3 minutes.
After three minutes, participants proceeded onto the third phase of the experiment, which was the exemplar testing phase. This was identical across all three learning conditions. Exemplar testing phase consists of two parts: testing of the studied items and testing of the novel items. Exemplar testing phase of experiments consisted of
recognition test. Therefore, in the testing phase, participants were presented with a word in the middle of the screen, and with the names of four categories from the learning phase and were asked to choose the right category for each word (See Figure4). For the
26
first part of the test, participants saw items from the encoding phase which can be classified as “old” items. Participants were asked to categorize them in one of the four categories. Participants were asked to press first letter of correct category name from the keyboard (e.g. V for vomidor). The testing phase was self-paced. The second part of the testing phase included only novel exemplars of the categories (new exemplars) that the participants did not study during encoding phase. Thus, they had to infer the category information.
Figure 4. Sample for testing trial
The last part of experiment was explicit testing of category information. Participants were given a paper which had non-word category names on it, and they were asked to think about all exemplars of categories that they studied. They were asked to find actual category names by using common properties of exemplars. The last part of the
27 2.2 Results
Learning Phase. Alpha level was set to .05 for all analyses. %80 correct answer in
encoding phase was determined as a lower limit to include each data in the analyses. All participants successfully chose the correct category in blocked (M = .99, SD = .12), interleaved (M = .99, SD =.13) and semi-interleaved (M =.98, SD = .39) learning phases. Median of each participants’ reaction times for encoding phase was calculated, as the median analysis provides more accurate results than the mean analysis for reaction times in terms of finding central tendency. In addition, mean analysis is more affected by outliers and extreme scores (Leys, Ley, Klein, Bernard, & Licata, 2013). For the encoding part of the experiment, mean of medians for each condition was analyzed. In addition, reaction time for the first and the second presentation of target word was analyzed separately. The mean of median RTs was submitted to a mixed analysis of variance (ANOVA), with the presentation time (first vs second) as the repeated measure and the learning condition as the between-subjects variable (blocked, interleaved and semi-interleaved learning conditions). There was a significant main effect of
presentation time (first vs second) on reaction time, F(1,99) = 152.58 , p < .001, MSE = 36666.22, ηp2 = .61. Participants got faster from first presentation (M = 2438, SD = 606.29) to second presentation (M =2107, SD = 687.12). There was a main effect of learning condition on reaction times (RT), F(2,99) = 22.46 , p < .001, MSE =
12580075.59, ηp2 = .31. Therefore, a Bonferroni post-hoc analysis revealed a significant difference between blocked (M = 1768, SD = 437.84) and interleaved learning
conditions (M = 2429, SD =552.50), p < .001, blocked and semi-interleaved (M = 2603,
28
between interleaved and semi-interleaved learning conditions in terms of reaction times to target stimuli, p = .53.
Table 2.Reaction times for first and second presentations of exemplars in learning phase
Reaction Time First
Presentation
Second Presentation
Total
Presentation
Mean(SD) Mean(SD) Mean(SD)
Blocked Condition 2026(417.18) 1547(510.04) 1786(437.79)
Interleaved Condition 2558(566.50) 2300(561.50) 2437(553.87)
Semi-interleaved Condition
2731(593.21) 2475(606.92) 2601(566.44)
There was a significant interaction between presentation RT and learning conditions,
F(2,99) = 7.64 , p = .001, MSE = 36666.22, ηp2 = .13. Post-hoc tests using pairwise comparisons were conducted to find reaction time (See Table2) differences between first and second presentation of category exemplars within participants. First and second RTs for each condition was separately submitted to a repeated measures ANOVA. In blocked condition, there was a significant difference between first presentation (M = 2026, SD = 417.18) and second presentation (M = 1547, SD = 510.04) of exemplars, F(1,33) = 76.96, p <.001, ηp2= .70. In interleaved condition, there was a significant reaction time difference between first (M = 2558, SD = 566.50) and second presentation (M = 2300,
29
condition, again there was a statistically significant difference between first presentation (M = 2731, SD = 101.73), and second (M = 2475, SD = 104.08) presentation, F(1,33) = 33.30, p <.001, ηp2 = .50. Shortly, participants became faster in pressing key from the keyboard in the second presentation of exemplars in all learning conditions.
Testing phase. For testing part of the experiment, testing accuracy of participants,
response times to target exemplars, study status (old vs. new exemplars) and category inferences were analyzed. First of all, accuracy of exemplars in testing was submitted to mixed analysis of variance (ANOVA) with study status (old vs. new exemplars) as repeated measure variable and the learning conditions (blocked, interleaved and semi-interleaved) as the between-subject variable. First, there was a significant main effect of study status on testing accuracy, F(2,99) = 35.37, p < .001, MSE = .01, ηp2 = .26,. Also, there was a significant main effect of learning condition on testing accuracy, F(2,99) = 5.26, p = .007, MSE = .09, ηp2 = .10. A Bonferroni post-hoc test showed that there was no statistically significant accuracy difference between blocked learning condition (M =.49, SD = .23) and interleaved learning conditions (M = .50, SD = .21), p = 1.00. On the other hand, there was significant difference between blocked (M =.49, SD = .23) and semi-interleaved (M = .64, SD = .21), p = .01 and between interleaved (M = .50, SD = .21) and semi-interleaved conditions, p = .03.
For testing accuracy, a two-way ANOVA analysis revealed that there was a significant interaction between study status (old vs. new) and learning conditions (blocked,
interleaved and semi-interleaved), F(2,99) = 4.24, , p = .017, MSE = .01, ηp2 = .08,. Moreover, post-hoc tests using pairwise comparison were conducted to find accuracy differences between for old and new testing exemplars within each learning condition
30
(See Table3). The accuracy for exemplars was submitted to a repeated measures
ANOVA separately for each learning condition, with the exemplar type (old vs. new) as the repeated measure. Results showed there was no significant difference between old blocked exemplars (M = .50, SD = .24) and new blocked exemplars (M = .48, SD =.23),
F(1,33) = 1.32, p = .26, ηp2 = .04. There was a significant difference between old (M = .54, SD = .22) and new exemplars (M = .47, SD = .21) for interleaved learning condition,
F(1,33) = 14.74, p < .001, ηp2 = .31. Lastly, there was significant difference between old (M = .69, SD = .22) and new exemplars (M = .59, SD = .22) for semi-interleaved
condition, F(1,33) = 31.93 , p < .001, ηp2 = .49 .
Table 3. Mean accuracy rates for old and new exemplars in exemplar testing phase and the standard deviations in parentheses
Accuracy Rates Old Exemplar New Exemplar
Mean(SD) Mean(SD) Blocked Condition .50(.24) .48(.23) Interleaved Condition .54(.22) .47(.21) Semi-interleaved Condition .69(.22) .59(.22)
Two different types of response times were analyzed separately: unconditional response times and response times conditionalized on accuracy. Unconditional analysis included all response times for all exemplars, regardless of whether participants identified the category correctly and incorrectly. For conditional analyses, response times only for
31
trials in which participants identified the categories correctly were included in the analyses. For testing unconditional response times, the mean of response times was submitted to a mixed ANOVA with study status (old vs. new exemplars) as within-subject variable and learning conditions (blocked, interleaved and semi-interleaved conditions) as between-subjects variable. Analysis showed that there was a main effect of study status (old vs new) on response times. Therefore, there was a statistically significant difference between old exemplars (M = 3291, SD = 1281.35) and new exemplars (M = 2716, SD = 987.43) for unconditional response times in testing phase
F(1,99) = 47.07, p < .001, MSE = 358600.54 ,ηp2 = .32, (See Table 4 for response times). There was not significant main effect of learning condition on response times,
F(2,99) = 2.33, p = .10, MSE = 5114730.38, ηp2 = .04,. Likewise, there was no significant interaction between study status (old vs. new) and learning conditions (blocked, interleaved &semi-interleaved) on response times, F(2,99) = 2.45 , p = .09,
MSE = 358600.54, ηp2 = .05.
Table 4.Unconditonal response times for new and old exemplars in exemplar testing part
Response Time Old Exemplar New Exemplar
Mean(SD) Mean(SD) Blocked Condition 3551(1415.64) 2797(1106.97) Interleaved Condition 3476(1272.09) 2824(1044.08) Semi-interleaved Condition 2847(1050.14) 2527(783.16)
32
In conditional analysis, the analyses were conducted in the same way, only analyzing response times for correct answers: the analyses revealed similar results. There was significant main effect of study status on conditional response times, F(1,99) = 31.67, p < .001, MSE = 433322.85, ηp2 = .24,. Therefore, there was a statistically significant difference between old exemplars (M = 3041, SD = 1159.62) and new exemplars (M = 2523.13, SD = 900.89) for conditional response times in testing phase. On the other hand, there was no significant main effect of learning conditions on conditional response times (See Table5), F(2,99) = 2.74, p = .07, MSE = 1660674.10, ηp2 = .05, . There was no significant interaction between study status (old vs. new) and learning conditions (blocked, interleaved & semi-interleaved), F(2,99) = 1.61 , p = .20, MSE = 433322.85, ηp2 = .03.
Table 5.Conditonal response times for new and old exemplars in exemplar testing part
Reaction Time Old Exemplar New Exemplar
Mean(SD) Mean(SD) Blocked Condition 3320 (1336.24) 2579 (936.04) Interleaved Condition 3149 (1090.75) 2676 (999.37) Semi-interleaved Condition 2656 (946.41) 2313 (732.19)
Last analysis was conducted to find participants’ inferences about categories and exemplars. Each participants’ inference for each category was coded by three coders. Participants’ inferences that were participants’ guesses about actual category names for studied categories were coded as “1” if they could infer the category name correctly and
33
“0” if the response was incorrect or the response was empty. In addition, response was coded as “.5” participant could understand the concept partially. For example, URTESA was non-word category name for “something that has round shape” and responses like “something that does not have an edge” were considered as “.5” point. After all categories were coded, participants’ scores were summed, and their proportion were calculated by dividing the sum to 12 (number of categories used in the study).
Table 6.Inference rate for actual category names
Inference rate
Mean(SD)
Blocked Condition .32(.26) Interleaved Condition .23(.27) Semi-interleaved Condition .38(.27)
A one-way ANOVA analysis conducted to find category inferences difference between learning conditions. Analysis revealed that there was not main effect of learning
conditions on category inferences F(1,99) = 2.90, p = .06, ηp2 = .05. Therefore, there was not significant differences between blocked, interleaved and semi-interleaved study conditions in terms of finding actual category names for exemplars (See Table6).
34
CHAPTER III
EXPERIMENT 2
3.1 Method 3.1.1 Participants
A hundred and two Bilkent University students same as in Experiment 1, who are between the ages of 18-30 and native speakers of Turkish participated in the study. Participants were compensated with course credit.
3.1.2 Material, Design and Procedure
Material, design and procedure were same as in Experiment 1 with one major difference. In Experiment 1, words were used as experimental material. Differently from
Experiment 1, Experiment 2 presented participants with pictures of words. The design and the procedure were identical to Experiment 1 except for the nature of the material. 3.2 Results
Learning Phase. All participants successfully chose the correct category in blocked (M =
.98, SD = .04), interleaved (M = .97, SD =.03) and semi-interleaved (M =.99, SD = .01) learning phase. The mean of median RTs was submitted to a mixed analysis of variance (ANOVA), with the presentation time (first vs second) as the within-subject measure and the learning condition as the between-subjects variable (blocked, interleaved and semi-interleaved learning conditions). There was a significant main effect of
35
36904.96, ηp2 =.68,. Participants got faster from first presentation (M = 2063.43, SD = 649.94) to second presentation (M =1675, SD = 635.99). There was a main effect of earning condition on RT, F(2,99) = 13.38 , p < .001, MSE = 607482.12, ηp2 = .22,. A Bonferroni post-hoc analysis was conducted that there was a significant difference between blocked (M = 1480, SD = 94.52) and interleaved learning conditions (M = 2182.07, SD = 513.53), p < .001, blocked (M = 1480, SD = 628.10) and semi-interleaved (M = 1945, SD = 501.86), p = .002. On the other hand, there were no significant reaction time differences between interleaved and semi-interleaved learning conditions, p = .24.
Table 7. Reaction times for first and second presentations of exemplars in learning phase
Reaction Time First Presentation Second Presentation
Total
Presentation
Mean(SD) Mean(SD) Mean(SD)
Blocked Condition 1654 (628.82) 1306 (659.12) 1480 (629.00)
Interleaved Condition 2254 (538.56) 2109 (509.72) 2182 (513.53)
Semi-interleaved Condition
2281 (590.15) 1609 (452.85) 1945 (501.86)
There was a significant interaction between presentation reaction times and learning conditions, F(2,99) = 32.48 , p < .001, MSE =36904.96, ηp2 = .40,. Post-hoc pairwise comparisons were conducted to find reaction times difference between first and second presentation of category exemplars separately for each learning condition to gain insight
36
about the interaction (See Table7). RTs for first and second presentations were entered into a repeated measures ANOVA as the repeated factor. Separate analyses for each learning condition were conducted. In blocked condition, there was a significant difference between first presentation (M =1654, SD = 628.82) and second presentation (M = 1306, SD = 659.12) of exemplars, F(1,33) = 53.20, p <.001, ηp2 = .62. In
interleaved condition, there was a significant difference between first (M = 2254, SD = 538.56) and second presentation (M = 2109, SD = 509.72) for reaction times, F(1,33) = 15.97, p <.001, ηp2 = .37 . In addition, in semi-interleaved learning condition, there was a significant difference between first presentation (M = 2281, SD = 590.15), and second presentation (M = 1609, SD = 452.85), F(1,33) = 154.23, p <.001, ηp2 = .83.
Testing phase. For testing part of the experiment, testing accuracy of participants,
response times to target exemplars, study status (old vs. new exemplars) and category inferences were analyzed. Accuracy of exemplars in testing was submitted to mixed analysis of variance (ANOVA) with study status (old vs. new) as within-subject variable and the learning condition (blocked, interleaved and semi-interleaved) as the between-subjects variable. (See Table8). First, there was a significant main effect of study status on testing accuracy, F(2,99) = 95.10, p < .001, MSE = .01, ηp2 = .50,. Also, there was a significant main effect of learning condition on testing accuracy, F(2,99) = 11.01, p < .001, MSE = .05, ηp2 = .18. A Bonferroni post-hoc test revealed that there was not a significant difference between blocked (M = .66, SD = .16) and interleaved learning condition (M = .63, SD = .19), p = 1.00. On the other hand, there was a significant accuracy difference between blocked and semi-interleaved (M = .80, SD = .14) learning conditions, p =.01 and between interleaved and semi-interleaved learning conditions, p <
37
.001. For testing accuracy a mixed ANOVA analysis revealed that there was not a significant interaction between study status (old vs. new) and learning conditions (blocked, interleaved and semi-interleaved), F(2,99) = 2.42, , p = .09, MSE = .01, ηp2 = .05.
Table 8. Mean accuracy rates for old and new exemplars in exemplar testing phase
Accuracy Rates Old Exemplar New Exemplar
Mean(SD) Mean(SD) Blocked Condition .69(.19) .62(.15) Interleaved Condition .68(.21) .57(.18) Semi-interleaved Condition .86(.14) .75(.14)
For response times to the exemplars, conditional and unconditional analyses were conducted like in Experiment 1. For testing unconditional reaction times, a two-way ANOVA analysis was conducted. Analysis failed to show main effect of study status (old vs new), there was not a statistically significant difference between old exemplars and new exemplars for unconditional response times in testing phase F(1,99) = .001, p = .98, MSE = 240601.85, ηp2 = .00, . There was a significant main effect of learning condition on response times, F(2,99) = 9.60, p < .001, MSE = 1027433.75, ηp2 = .16. A Bonferroni post-hoc analysis was conducted that, there was not a significant difference between blocked (M = 2586, SD = 580.24) and interleaved learning conditions (M
38
=2744, SD =960.99), p = .10, but there was a significant difference between blocked and semi-interleaved strategy (M =2148, SD = 530.06) for first and second presentation reaction times, p < .001. There was a significant difference between interleaved and semi-interleaved learning conditions in terms of reaction times to target stimuli, p = 003. There was significant interaction between study status (old vs. new) and learning
conditions (blocked, interleaved & semi-interleaved) on response times, F(2,99) = 5.07,
p = .01, MSE = 240601.85, ηp2 = .10,. Also, post-hoc test using pairwise comparison was conducted to find was conducted to find response time (See Table9) difference between old and new exemplars through a repeated measures ANOVA, separately for each learning condition. In blocked condition, there was a significant difference between old exemplars (M = 2983, SD = 651.51) and new exemplars (M = 2729, SD = 683.49) of exemplars, F(1,33) = 5.02, p = .03, ηp2 = .13. In interleaved condition, there was not a significant response time difference between old (M = 2755, SD = 1072.74) and new exemplars (M = 2734, SD = 994.08), F(1,33) = .03, ηp2 = .001, p = .88. In addition, in semi-interleaved learning condition, there was a statistically significant difference between old exemplars (M = 2007, SD = 600.90), and new exemplars (M = 2288, SD = 642.21), F(1,33) = 6.33, p = .02, ηp2= .16.
Table 9.Unconditional response times for new and old exemplars
Response Time Old Exemplar New Exemplar
Mean(SD) Mean(SD)
Blocked Condition 2983 (651.51) 2729 (683.49)
